Spatial Regression in Health: Modelling Spatial Neighbourhood of High Risk Population

摘要

Many health conditions affect certain individuals more than others: for example, adults over 65 years of age are more affected by cardiovascular disease than younger individuals. Therefore, the spatial pattern of the disease incidence can be modelled more effectively through the residential pattern of higher risk groups. The method is demonstrated through a spatial regression of the association of cardiac catheterization and socioeconomic determinants in Calgary (Canada). Over a 5-year interval, 45% of catheterizations are performed on seniors, that constitute 9% of the population. Seniors' residential location is therefore used as an auxiliary process to model the spatial weights of the regression model. This spatial model leads to a more realistic neighbourhood configuration, yielding more reliable regression estimates. Based on the residential location of the population at greater risk, the model presents low sensitivity to variations in the supporting geographic units. The use of a relevant auxiliary process is general and applicable to a range of conditions; it constitutes a promising alternative to the direct estimation of spatial parameters on the primary process. Overall, the spatial weights matrix based on at risk population shall increase the reliability of spatially autoregressive multivariate epidemiological models.